The artificially generated media, popularly known as deepfakes, has become a major threat to data integrity and information. This type of content can be used to mislead the masses by manipulating videos of influential personalities to spread misinformation. The research addresses this challenge by proposing a system for its detection, which combines the strengths of convolutional and recurrent networks, recovering features from individual video frames, which exposes minute irregularities generated by various manipulation techniques. These extracted features are then fed into the LSTM network, which is effective at analyzing temporal sequences and can detect subtle discrepancies in the progression of facial expressions and movements in the video frames. The model provides an accurate and efficient way to detect deepfakes and contributes to the improvement of traditional techniques in the process.